Overview of Artificial Intelligence

Overview of AI

• Artificial Intelligence (AI) is a computer science field focused on creating computers or machines as intelligent as humans.
• John McCarthy, a pioneer in AI, defined it as "the science and engineering of making intelligent machines, especially intelligent computer programs."
• AI aims to make computers, robots, or software think intelligently, similarly to humans.
• AI is achieved by studying the human brain's thought processes and how humans learn, decide, and problem-solve, using these findings to develop intelligent software and systems.
• Early AI development was driven by the question "Can a machine think and behave like humans?"
• AI's two primary goals: create expert systems that exhibit intelligent behavior and implement human intelligence in machines to enable understanding, thinking, learning, and behaving like humans.

Contributing Disciplines to AI

• AI draws from disciplines like computer science, biology, psychology, linguistics, mathematics, and engineering.
• Focuses on developing computer functions associated with human intelligence, such as reasoning, learning, and problem-solving.

Programming With and Without AI

• AI programs can handle generic questions, while non-AI programs are limited to specific queries.
• AI programs can easily incorporate new modifications by integrating independent pieces of information, allowing modifications without affecting the overall structure.
• Non-AI programs require structural changes for modifications, which can be difficult and may adversely affect the program.

AI Technique Essentials

• AI techniques should handle large volumes of knowledge.
• Organized and well-formatted.
• Constantly evolving.
• AI techniques should make knowledge perceivable to providers, easily modifiable for error correction.
• Useful across various situations even when incomplete or inaccurate.
• AI techniques should enhance the speed of execution for complex programs.

AI Applications

• Gaming: Essential for strategic thinking in games like chess and poker, enabling machines to evaluate numerous possible positions using heuristics.
• Natural Language Processing: Enables computers to comprehend natural human language.
• Expert Systems: Integrates machines, software, and specialized information for reasoning and advising, providing explanations and advice to users.
• Vision Systems: Interpret and comprehend visual input, used in applications like aerial imagery analysis, medical diagnoses, and facial recognition by law enforcement.
• Speech Recognition: Intelligent systems can understand spoken language, including variations in accents and background noise.
• Handwriting Recognition: Software that converts handwritten text into editable digital text.
• Intelligent Robots: Capable of performing tasks, detecting physical data via sensors, using processors efficiently, exhibiting intelligence, learning from mistakes, and adapting to new environments.

History of AI (20th Century)

• 1923: Karel Čapek's play "Rossum's Universal Robots" (RUR) coined the word "robot."
• 1943: Foundations for neural networks were established.
• 1945: Isaac Asimov coined the term "Robotics".
• 1950: Alan Turing introduced the Turing Test for intelligence evaluation and published "Computing Machinery and Intelligence."
• 1956: John McCarthy coined the term "Artificial Intelligence." The first AI program was demonstrated at Carnegie Mellon University.
• 1958: John McCarthy invented the LISP programming language for AI.
• 1964: Danny Bobrow's dissertation at MIT demonstrated that computers could understand natural language to solve algebra problems.
• 1965: Joseph Weizenbaum at MIT developed ELIZA, an interactive program that carries on a dialogue and mimics human conversation in English.
• 1969: Scientists at Stanford Research Institute developed Shakey, a robot with locomotion, perception, and problem-solving abilities.
• 1973: The Assembly Robotics group at Edinburgh University created Freddy, a robot capable of using vision to locate and assemble models.
• 1979: The first computer-controlled autonomous vehicle, the Stanford Cart, was built.
• 1985: Harold Cohen created Aaron, a drawing program.
• 1990: Significant advances in machine learning, case-based reasoning, multi-agent planning, scheduling, data mining, web crawlers, natural language understanding, virtual reality, and game playing.
• 1997: The Deep Blue Chess Program defeated Garry Kasparov, the world chess champion.
• 2000: Interactive robot pets became commercially available; MIT introduced Kismet, while Nomad explored Antarctica.

Intelligent Systems - Defining Intelligence

• Intelligence: A system's ability to calculate, reason, perceive relationships, learn from experience, store and retrieve information, solve problems, comprehend complex ideas, use natural language fluently, classify, generalize, and adapt to new situations.

Types of Intelligence

• Linguistic intelligence: Speak, recognize, and utilize phonology (speech sounds), syntax (grammar), and semantics (meaning)
• Musical intelligence: Create, communicate with, and understand meanings made of sound, understanding of pitch and rhythm
• Logical-mathematical intelligence: Use and understand relationships in the absence of action or objects, understanding complex and abstract ideas
• Spatial intelligence: Perceive visual/spatial information, change it, and re-create visual images without reference to objects, create 3D images, and move/rotate them
• Bodily-Kinesthetic intelligence: Use complete or part of the body to solve problems or fashion products, control fine/coarse motor skills, and manipulate objects
• Interpersonal intelligence: Recognize and make distinctions among other people's feelings, beliefs, and intentions
• Intrapersonal intelligence: Distinguish among one's feelings, intentions, and motivations

Components of Intelligence

• Reasoning: Processes enabling judgment, decision-making, and prediction.
  • Inductive Reasoning: Derives broad general statements from specific observations.
  • Deductive Reasoning: Examines possibilities from a general statement to reach a specific, logical conclusion.
• Learning: Gaining knowledge/skill via study, practice, teaching, or experience, enhancing awareness.
  • Auditory Learning: Through listening and hearing.
  • Episodic Learning: Remembering sequences of events.
  • Motor Learning: Precise muscle movement.
  • Observational Learning: Imitating others.
  • Perceptual Learning: Recognizing stimuli.
  • Relational Learning: Differentiating stimuli based on relationships rather than absolutes.
  • Spatial learning: Using visual stimuli.
  • Stimulus-Response Learning: Performing behavior upon stimulus.
• Problem Solving: Perceiving and achieving a solution from a current state, overcoming obstacles.
• Decision Making: Selecting the optimal alternative from multiple options to achieve a goal.
• Perception: Acquiring, interpreting, selecting, and organizing sensory information.
• Linguistic Intelligence: Comprehending and using both verbal and written language effectively.

Human vs Machine Intelligence

• Humans perceive through patterns whereas machines perceive through rules and data.
• Humans recall with patterns, but machines use search algorithms.
• Humans can extrapolate from incomplete data, but machines struggle with incompleteness.

Research Areas of AI

• Speech and Voice Recognition: Both relate to spoken input but differ, focusing on understanding what is spoken (speech) versus identifying who is speaking (voice); speech recognition is not speaker-dependent and voice recognition needs explicit training..

Real Life AI Applications

Application: Expert Systems

• Sr. No. 1
• Research Area: Expert Systems
• Real Life: Flight-tracking systems, clinical systems

Application: Natural Language Processing

• Sr. No. 2
• Research Area: Natural Language Processing
• Real Life: voice recognition, Automatic voice output, Google Now feature

Application: Neural Networks

• Sr. No. 3
• Research Area: Neural Networks
• Real Life: handwriting recognition, character recognition, Pattern recognition systems such as face recognition,

Application: Robotics

• Sr. No. 4
• Research Area: Robotics
• Real Life: Industrial robots for moving, spraying, painting, precision checking, drilling, cleaning, coating, carving etc.

Application: Fuzzy Logic

• Sr. No. 5
• Research Area: Fuzzy Logic
• Real Life: automobiles, Consumer electronics, etc.

Task Classification in AI

• AI tasks are categorized as Formal, Mundane, and Expert tasks.

Task Domains of Artificial Intelligence

• Mundane (Ordinary) Tasks:
  • Perception: Computer Vision, Speech, Voice
  • Natural Language Processing: Understanding, Language Generation, Language Translation
  • Common Sense, Reasoning, Planning
  • Robotics: Locomotive
• Formal Tasks:
  • Mathematics: Geometry, Logic, Integration and Differentiation
  • Games: Go, Chess (Deep Blue), Checkers
  • Verification, Theorem Proving
• Expert Tasks:
  • Engineering, Fault Finding, Manufacturing, Monitoring
  • Scientific Analysis, Financial Analysis, Medical Diagnosis
  • Creativity
• Humans acquire mundane tasks first; AI now prospers more in expert tasks due to knowledge representation demands of mundane tasks.

Agents and Environments – Defining Agents

• An AI system is made of an agent and its environment, agents functioning within the environment (which may contain other agents).
• An agent is something that perceives its environment via sensors and acts upon it via effectors.
  • A human agent has sensory organs (eyes, ears) as sensors and organs (hands, legs) as effectors.
  • A robotic agent uses cameras/infrared as sensors, motors/actuators as effectors.
  • A software agent uses encoded bit strings as programs and actions.
• Agents Terminology:
  • Performance Measure: Criteria determining agent success.
  • Behavior of Agent: Actions performed after given percepts.
  • Percept: Agent's perceptual inputs.
  • Percept Sequence: History of what an agent has perceived.
  • Agent Function: Maps precept sequence to actions.
• Rationality: Judgement or being reasonable. Concerned with actions and outcomes depending on perceived variables. Important for obtaining useful infromation.
• Ideal Rational Agent's actions on basis of:
  • Its percept sequence
  • Its built-in knowledge base
• The aspects to be determined for rationality:
  • Performance measures
  • Percept Sequence
  • Knowledge of environment
  • Available actions
• Rational agent performs right action. Problem characterized by Performance Measure, Environment, Effectors, and Sensors.
• Structure of Intelligent Agents: Agent = Architecture + Agent Program, with architecture being the machinery and the program being the implementation.
• Simple Reflex Agents choose actions based only on the current percept, are rational only if the decision is based on current percept, and need the environment to be completely observable.
• Model-Based Reflex Agents: Maintain an internal state to choose actions, using a model for how things happen.
• Goal-Based Agents choose actions to achieve a goal.
• Utility-Based Agents choose actions based on preference (utility) of each state.

Defining Environments

• Some programs operate in an artificial environment confined to database, keyboard and computer file systems
• Other software robots exist in a rich, unlimited environmental simulation
• The Turing Test environment assesses "intelligent behavior”, and involves a human tester distinguishing typed responses from machine vs human
• This tests the computer's intelligence

Properties of the Environment

• Discrete or Continuous: States are defined, or they are continuous
• Observable or Partially Observable: Environment can be observed, or not
• Static or Dynamic: Environment changes while acting, or not
• Single agent or Multiple agents: May contain other agents, or not
• Accessible vs. inaccessible: Agent access to the complete state, or not
• Deterministic vs. Non-deterministic: Next state completely determined, or not
• Episodic vs. Non-episodic: Quality of action depends just on the episode itself, or not

Popular Search Algorithms – Fundamentals

• Searching is a technique for problem solving in AI, search algorithms help locate a particular location.
• Single Agent Pathfinding Problems: E.g. 3X3 eight-tile, 4X4 fifteen-tile, and Travelling Salesman
• Problem Space: The environment in which the search takes place.
• Problem Instance: Initial state + Goal state
• Problem Space Graph: Graph that represents the problem state with nodes and operators as edges
• Depth of a problem: Shortest path length.
• Space Complexity: Max nodes stored in memory.
• Time Complexity: Max nodes created.
• Admissibility: Algorithm to find an optimal solution.
• Branching Factor: Average child nodes in problem space graph.

Different Search Strategies

Brute-Force Search Strategies

• Simplest, lacking domain-specific knowledge, effective with few states.

• State description needed.

• Requirements: Valid Operators, Initial/Goal state descriptions.

• Breadth-First Search:

  • Explores neighboring nodes first.
  • Uses FIFO queue.
  • Provides shortest path.
  • Disadvantage: High memory usage (exponential)

• Depth-First Search:

  • Uses recursion with LIFO stack.
  • Implemented differently.
  • Disadvantage: May not terminate. Bidirectional Search: Search from forward and backward, joining in the middle.

• Uniform Cost Search: Sorts by increasing path cost; like Breadth-First if costs are equal.

• Iterative Deepening Depth-First Search: Depth-first searches to level 1, then 2, etc., until solution.

Informed (Heuristic) Search Strategies

• Heuristic Evaluation Functions estimate cost of optimal path, e.g., in sliding-tiles, count moves from goal state
• Pure Heuristic Search expands nodes in order of heuristic values, storing expanded and unexpanded nodes.
• The A Search (A-Star Search) avoids expensive paths, expanding promising paths first, based on value.
• Best-known form of Best Search.

Where:

• f(n) = g(n) + h(n)
• g(n) is is the cost to reach the node
• h(n) is the estimated cost to get from the node to the goal
• f(n) is estimated total cost of path through n to goal
• Greedy Best First Search expands closest node based on f(n) = h(n), using priority queue

Local Search Algorithms

• They start from a prospective solution and then move to a neighboring solution.
• They can return a valid solution even if it is interrupted at any time before they end.
• Hill Climbing: Starts with arbitrary solution, incrementally improves until no further improvement.
• Local Beam Search: Holds k states, creates successors to find maximum, sorts, and picks highest k.
• Simulated Annealing: Allows acceptance of changes under variation.
• Traveling Salesman Problem: Finds lowest-cost tour visiting each city.

Fuzzy Logic Systems (FLS)

• Fuzzy Logic Systems (FLS) produce acceptable output in response to incomplete, ambiguous, distorted, or inaccurate (fuzzy) input.

What is Fuzzy Logic?

• Fuzzy Logic (FL) is an approach to reasoning resembling human reasoning. Involves all intermediate possibilities between digital YES and NO values
• Works on "degrees of truth" rather than binary logic
• Inventor Lotfi Zadeh observed decision-making includes a range of possibilities
• Implemented using small micro-controllers and even workstation based control systems
• FLS is useful for commercial and engineering applications and can control consumer products

Fuzzy Logic Systems Architecture

• Fuzzification Module transforms system inputs and crisp numbers into fuzzy sets, splitting input signal into five steps
• Knowledge base stores IF-THEN rules provided by experts.
• Inference Engine simulates the human reasoning process by making fuzzy inference on the inputs and IF-THEN rules.
• Defuzzification Module transforms the fuzzy set obtained by the inference engine into a crisp value.
• Membership functions work on fuzzy sets of variables, graph fuzzy sets, quantifiy linguistics
• Universe of discourse (X)
• The triangular membership function shapes are most common among various other functions

Fuzzy Logic Example

• An example 5-level air conditioning system with a fuzzy logic system adjusts the temperature by comparing the room temperature and target temperature values.

Key Algorithm Steps

1. Define linguistics (start)
2. Construct membership functions (start)
3. Construct knowledge base of rules (start)
4. Convert crisp data into fuzzy data sets using membership functions (fuzzification)
5. Evaluate rules in the rule base (inference engine)
6. Combine results from each rule (inference engine)
7. Convert output data into non-fuzzy values. (defuzzification)

FLS Application

• Automotive Systems uses automatic gearboxes
• Four-Wheel and Vehicle environment control
• Consumer Electronics uses Hi-Fi Systems and televisions
• Domestic Goods uses microwave ovens
• Enivornment control uses air conidtioners

Advantages and Disadvantages

Advantages

Mathematical concepts within fuzzy reasoning are very simple. You can modify a FIS by just adding or deleting rules due to flexibility of fuzzy logic. Fuzzy logic Systems can take imprecise, distorted, noisy input information. FLSs are easy to construct and understand. Fuzzy logic is a solution to complex problems in all fields of life, including medicine, as it resembles human reasoning and decision making.

Disadvantages

There is no systematic approach to fuzzy system designing. They are understandable only when simple. They are suitable for the problems which do not need high accuracy.

Natural Language Processing (NLP) – Definition

• NLP: AI method for machines to communicate using natural language like English, which is required when you want an intelligent system to perform what you want
• NLP makes computers perform tasks using natural languages which can be speech and text
• NLP involves two components: Natural Language Understanding and Generation

Natural Language Components

• NLU Tasks: Maps natural language input into representations and analyzes aspects.
• NLG Process: Produces meaningful natural language phrases/sentences from internal representation including text and sentence planning

Challenges in Natural Language Processing

• NL is ambiguious and structured
  • Lexical ambiguity: At word-level
  • Syntax Level ambiguity: A sentence can be parsed in different ways
  • Referential ambiguity: Referring to something using pronouns
• In addition:
  • One input can mean different things
  • Many inputs can mean the same thing

NLP Key Terminology

• Phonology: Systematically organizing sound.
• Morphology: Constructing words units.
• Morpheme: Primitive unit of meaning.
• Syntax: Arranging words in sentences.
• Semantics: Word meanings.
• Pragmatics: How context impacts understanding.
• Discourse: Impact of preceding sentences.
• World Knowledge: General information.

Process Steps and Grammar

• Lexical Analysis: Text broken down into paragraphs, sentences, and words.
• Syntactic Analysis: Grammar analysis, phrase structure
• Semantic Analysis: Meaning of the text.
• Discourse Integration: Effect of surrounding sentences.
• Pragmatic Analysis : Re-interpreting the user based on the actual meant

Analyzing Syntactic Structures

• Algorithms for syntactic analysis involve:
  • Context Free Grammar (Most common type)
  • Top-Down Parser

Expert Systems Overview

• Expert Systems (ES): Computer-applications solving complex, domain-specific problems at a level of extra-ordinary human intelligence.
• Stanford University created ES
• ES should be high-performance, understandable, reliable, and responsive.
• Includes knowledge base, inference engine, and user interface

Capabilities of Expert Systems

• Capable of: advising, decision assistance, demonstrating, diagnostics, explanation
• Incapable of making decisions

Expert System Components

• Knowledge Base: Contains high-quality domain-specific information, success that is dependent on collection of highly accurate and precise knowledge.
• Inference Engine: Derives solutions via rules, using forward or backward chaining.
  • The inference engine uses efficient procedures and rules for deductions
• User Interface: Provides natural language interaction, explaining reasoning in natural language..

Efficient User Interface

• Users should accomplish goals, work to meet existing/desired practices.
• It should be adaptable.
• Limit use of user input.

Production of Knowledge

• Use knowledge representations (such as IF-THEN), and have acquisition process via researchers/engineers

Expert Systems Limitations

• Requires computer/development resources, with limitations in understanding and knowledge

The table below shows where ES can be applied.

Application | Description

• • | -- Design Domain | Camera lens design, automobile design. Medical Domain | Diagnosis Systems to deduce cause of disease from observed data, conduction medical operations on humans. Monitoring Systems | Comparing data continuously with observed system or with prescribed behavior such as leakage monitoring in long petroleum pipeline. Process Control| Systems Controlling a physical process based on monitoring. Knowledge Domain | Finding out faults in vehicles, computers. Finance/Commerce | Detection of possible fraud, suspicious transactions, stock market trading, Airline scheduling, cargo scheduling.

Technology to build systems

• Tools should be able to reduce the effort required
• Shells should help knowledge developers for knowledge acquisition

Implementation Steps

• 1. Identify Problem Domain
• 2. Design the System
• 3. Develop the Prototype
• 4. Test and Refine the Prototype
• 5. Develop and Complete the ES
• 6. Maintain the System

Benefits of Expert Systems

• Availability: Software makes them available
• Less Production Cost: Cost is less, while speed of implementation is more
• Less Error Rate: Error is much lower
• Steady response: Constant without any change

Robotics Overview

• Robotics deals with the manipulation of objects for a variety of reasons
• Branch of AI composed of Electrical Engineering, and Computer Science for application of robots

Aspects of Robotics

• Mechanical construction or form
• Electrical Components
• Computer Program

Robot vs Al Programs

• Usually operate in computer- stimulated worlds vs real physical world
• The input to an Al program is in symbols and rules vs Inputs to robots is analog signal
• Al Program operate on general purpose computers vs Need special hardware with sensors and effectors

Robot Locomotion Overview

• Mechanism making robot capable to move.
• Various types of locomotions:
• Legged, Wheeled, Combination of Legged and Wheeled Locomotion

Wheeled Locomotion

• Requires less motors, is easy to implement and reliable, efficient and more reliable vs legged locomotion.

Components of Robots

• Power Supply
• Actuators
• Electric motors (AC/DC)

Computer Vision

• Al Technology that helps in security, safety, and entertainment.
• Hardware usually includes display device for monitoring the system

Computer vision: list of tasks

• OCR
• Face Detection

Application Domains

• autonomous vehicles
• biometrics

Use Case - Industries:

Handling material and welding

Neural Network - Overall Summary

• They are inspired from the natural system of the human brain.
• Basic Structure of ANNs is on the beliefs of the human brain, they can be imitated with wires. Made of living dendrites and neurons.
• The human brain is 100 billion cells called neurons, that connect to other cells. Accepted by dendrites
• Multiples nodes and biological neurons connect to imititate the human brain. This can take inut data and simplify, and is called activation
• Can learn by alltering values, by various weights

Artificial Neural Networks (ANNs) Topologies

• Types: FeedForward and Feedback
• Feedforward is unidirectional, used in pattern generation and fixed inputs
• Feedback has memory loops
• Use weights to indicate a connection that controls the signal between the two neurons.

Machine Learnings

• Supervised, where a teacher feeds the system information
• Unsupervised, no one knows what the outcome will be

Learning Strategies

• Reinforcement Learning is when a machine builds by observation
• Back propagation algorithms

Bayesian Networks

• Used in a a set of randon variables; beliefs of a machine
• Arcs should specify the direction
• They have conditions for discrete variables.

Applications of Neural Networks

• Networks carry out tasks that are easy for a human but difficult for a "machine": List of applications:
• Aerospace, Automotive
• Military, Electronics
• Financial, Industrial

AI Issues - Overall Concept

AI systems improve so quickly, they might even seem magical! This could be dangerous and cause people to be threatened

The Key Issues

• Privacy could be compromised
• Human Dignity could be sacrificed
• Safety could be at risk

AI Terminology

Term | Meaning

• • | -- Agent | Software programs with goals. Autonomous Robot | Robot free from external control. Backward Chaining | Working backward for a reason. Blackboard | Memory to interact. Environment | Part of the real world . Forward Chaining | Working to for a conclusion. Heuristics | Knowledge from experimentation. Knowledge Engineering | Knowledge from humans. Percepts | Format for infrmation. Pruning | Overriding considerations. Rule | Format for representing knowledge. Shell | Software for knowledge base. Task | Goals. Turing Test | Designed test the intelligence of a machine.